References
Research papers
Section titled “Research papers”- DINOv3 — satellite-pretrained encoder.
- Vision Transformers for Dense Prediction (DPT) — decoder design.
- AdaBins — adaptive depth bins.
- Depth Anything V2 — backbone ablation.
- GAMUS · SynRS3D — benchmark and synthetic training data.
- IM2HEIGHT · IM2ELEVATION — single-view height estimation.
Datasets and imagery
Section titled “Datasets and imagery”- GAMUS · SynRS3D
- US3D / DFC2019 · DFC2023 Track 2
- IARPA MVS3DM · NEON AOP LiDAR
- Cartosat sample products (NRSC Bhoonidhi)
- Copernicus DEM · SRTM · NASADEM · OpenStreetMap
See dataset details and packed training stores.
Models and weights
Section titled “Models and weights”Software and standards
Section titled “Software and standards”- Machine learning: PyTorch · Hugging Face Transformers · ONNX Runtime
- Geospatial: GDAL · Rasterio · PROJ
- Service and interface: FastAPI · Gradio · React · Three.js · Astro Starlight
- Build and UI: Vite · TypeScript · Mantine · React Three Fiber · ECharts · Vercel
For the project-specific setup and usage, see architecture, inference and software credits.